Research map / Label-efficient and robust learning / Semi-supervised learning
PAF: Perturbation-Aware Filtering for Open-Set Semi-Supervised Learning
Yinan Han, Qing-Yuan Jiang
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add perturbation-aware filtering of unlabeled data (ablation) | PAF · CIFAR-100 (mismatch 0.3) | seen-class accuracy (%): 71.2 -> 77.7 | Improved |
| use separate projection heads for consistency and rotation branches (ablation) | PAF · CIFAR-100 (mismatch 0.3) | seen-class accuracy (%): 61.8 -> 77.7 | Improved |
| add representation consistency loss (ablation) | PAF · CIFAR-100 (mismatch 0.3) | seen-class accuracy (%): 75.1 -> 77.7 | Improved |
| add auxiliary rotation prediction loss (ablation) | PAF · CIFAR-100 (mismatch 0.3) | seen-class accuracy (%): 75.6 -> 77.7 | Improved |
| use representation instability instead of confidence mutation for filtering (ablation) | PAF · CIFAR-100 (mismatch 0.3) | seen-class accuracy (%): 74.5 -> 77.7 | Improved |
| use representation instability instead of confidence mutation for filtering (ablation) | PAF · CIFAR-100 (mismatch 0.3) | OOD AUC (%): 78.2 -> 84.5 | Improved |
| use representation-level filter instead of confidence-based filter (ablation) | PAF · CIFAR-100 (mismatch 0.3) | ID sample retention rate and OOD filtering rate: +9.9% ID retention, +12.2% OOD filtering | Improved |
| increase perturbation trials Ka beyond 1 (ablation) | PAF · CIFAR-100 (mismatch 0.3) | AUC (%) | No clear effect |
| use Otsu adaptive threshold instead of fixed or percentile thresholds (ablation) | PAF · CIFAR-100 (mismatch 0.3) | seen-class accuracy and AUC | Improved |
| rescale Otsu threshold by gamma around 1.0 (ablation) | PAF · CIFAR-100 (mismatch 0.3) | seen-class accuracy and AUC | No clear effect |
| plug PAF into CFSG-CLIP | CFSG-CLIP · CUB-200-2011 (fine-grained OSSL) | seen-class accuracy (%): 84.73 -> 85.16 | Improved |
| use PAF instead of T2T for open-set SSL training cost | PAF · CIFAR-100 (mismatch 0.3) | training time (hours): 29.3 -> 23.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- ProSub — ProSub: Probabilistic Open-Set Semi-Supervised Learning with Subspace-Based Out-of-Distribution Detection (ECCV 2024) · better on 3 of 3 shared results
- SCOMatch — SCOMatch: Alleviating Overtrusting in Open-set Semi-supervised Learning (ECCV 2024) · better on 2 of 2 shared results
- CaliMatch — CaliMatch: Adaptive Calibration for Improving Safe Semi-supervised Learning (ICCV 2025) · better on 1 of 1 shared result
- OpenMatch — OpenMatch: Open-Set Semi-supervised Learning with Open-set Consistency Regularization (NeurIPS 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Let the Void Be Void: Robust Open-Set Semi-Supervised Learning via Selective Non-Alignment (AAAI 2026 · similarity 0.86)
- Targeted Representation Alignment for Open-World Semi-Supervised Learning (CVPR 2024 · similarity 0.85)
- Erasing the Bias: Fine-Tuning Foundation Models for Semi-Supervised Learning (ICML 2024 · similarity 0.84)
- Keep It on a Leash: Controllable Pseudo-label Generation Towards Realistic Long-Tailed Semi-Supervised Learning (NeurIPS 2025 · similarity 0.83)
- CRoFT: Robust Fine-Tuning with Concurrent Optimization for OOD Generalization and Open-Set OOD Detection (ICML 2024 · similarity 0.83)
- Revisiting Semi-Supervised Learning in the Era of Foundation Models (NeurIPS 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
